1. Field
The present technology relates to methods and apparatuses for improving optical proximity correction (OPC) model building, capable of extrapolation to any process condition within focus-exposure matrix (FEM), which can reduce the number of process combinations at which sample structures are made and empirically measured sample data are collected.
2. Description of Related Art
Conventional process window (PW) capable OPC models are calibrated at multiple process conditions (defocus and dose variations) across a focus-exposure matrix (FEM). However, such models cannot guarantee accurate prediction of new process conditions that are different from process conditions used in model calibration. The more process conditions that are included in the conventional OPC model calibration, the better the expected quality of the conventional process variation aware OPC model. However, multiple process conditions require additional wafer exposures and empirically measured data collection, and also increases the time needed to calibrate process variation aware OPC model.
With conventional process variation aware OPC model calibration, compact models are suited for predicting critical dimensions (CDs) only at the process condition(s) used in model calibration. When new process conditions vary from the process condition(s) used in model calibration, such models cannot guarantee correct prediction at the new process conditions (e.g., of the CDs at the new process conditions). Incomplete decoupling of the model's optical and resist effects disallows successful process window extrapolation beyond the process condition(s) used in model calibration.
FIG. 1 shows a conventional calibration flow. The OPC models consist of mask, optical, and resist components; and corresponding model parameters. An optical model is generated by changing film stack placement and exposure dose values 12. In calibration of the conventional OPC model, the model is calibrated at several process conditions 14 using a global optimization. A resist model is generated by adding resist effects 20. The model parameters are optimized during the model calibration process for the best possible match to sets of empirical data 22, 24, 26. Model calibration is performed until achieving an acceptable error tolerance of the modeled critical dimension 28. The result is the optimized OPC model 30.
However, the ability of the model to extrapolate to new process conditions other than those used for calibration, strictly depends on the number of process conditions included in model calibration. Also, significant time is required to calibrate such models, since the model calibration runtime is a linear function of the number of process conditions. Also, compact OPC models for 45 nm and smaller technology nodes have a large number of parameters that must be optimized during the model fitting process for the best possible match with the empirical data. Model calibration runtime has increased due to the complexity and increased number of model components and parameters to be optimized. Each process condition in the model calibration requires the collection of empirical data from exposed wafers made at the corresponding process condition, which leads to the additional requirement of collecting and analyzing significant quantities of process data. Also, hardware limits the number of process conditions used in a conventional PW OPC model calibration, since the physical computer memory usage is a linear function of the number of process conditions.
Traditional OPC models only compute the optimal pattern layout to optimize lithography patterning at the best process condition. An OPC model performs fast and reliable critical dimension (CD) prediction of all features present in the design layout. Accurate prediction of CD changes and lithographic effects under varying process conditions is beyond the scope of these mathematical models. OPC models that model only nominal process conditions are insufficient, due to inevitable process variation (such as defocus and dose variations) during production that place the post-OPC layout at a non-negligible patterning failure risk.